mirrors/scylladb
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

logalloc: increase capacity of _regions vector outside reclaim lock

Browse Source 
Reclaim consults the _regions vector, so we don't want it moving around while
allocating more capacity. For that we take the reclaim lock. However, that
can cause a false-positive OOM during startup:

1. all memory is allocated to LSA as part of priming (2baa16b371)
2. the _regions vector is resized from 64k to 128k, requiring a segment
   to be freed (plenty are free)
3. but reclaiming_lock is taken, so we cannot reclaim anything.

To fix, resize the _regions vector outside the lock.

Fixes #6003.
Message-Id: <20200311091217.1112081-1-avi@scylladb.com>
This commit is contained in:
Avi Kivity
2020-03-11 11:12:17 +02:00
committed by Tomasz Grabiec
parent 931d2fca45
commit c020b4e5e2
1 changed files with 11 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
utils/logalloc.cc
View File

@@ -2068,6 +2068,17 @@ bool segment_pool::migrate_segment(segment* src, segment* dst)
}
void tracker::impl::register_region(region::impl* r) {
// If needed, increase capacity of regions before taking the reclaim lock,
// to avoid failing an allocation when push_back() tries to increase
// capacity.
//
// The capacity increase is atomic (wrt _regions) so it cannot be
// observed
if (_regions.size() == _regions.capacity()) {
auto copy = _regions;
copy.reserve(copy.capacity() * 2);
_regions = std::move(copy);
}
reclaiming_lock _(*this);
_regions.push_back(r);
llogger.debug("Registered region @{} with id={}", r, r->id());